nulla 0.0.9

Nulla: a local AI companion bootstrapper (Windows) with voice (Whisper ASR + XTTS v2 TTS), llama.cpp + OpenHermes GGUF, and built-in mini-games.

Nulla

A local Windows bootstrapper for a talkative female elf AI companion — sets up Whisper (ASR), XTTS v2 (TTS), llama.cpp with OpenHermes GGUF, and ships sample mini-games.

---

Status

Alpha — functional CLI. nulla setup creates isolated venvs, fetches llama.cpp Windows binaries and the OpenHermes-2.5-Mistral-7B-GGUF Q8_0 from official upstreams, and wires in Whisper (CPU) + XTTS v2 (CUDA/CPU).
This package does not redistribute third-party models/binaries; they are downloaded during setup under their respective licenses.

Tested Environment

• GPU: RTX 5070 Ti 16 GB
• CPU: Ryzen 5 5600X
• RAM: 32 GB
• Storage: ~20 GB free recommended
• OS: Windows 11
• Python: 3.11.6 (required)

Requirements

• Windows 11
• Python 3.11.6 exactly
• NVIDIA CUDA (optional; CPU fallbacks exist but are slower, untested)

Python requirement: This project targets Python 3.11.6 exactly.

What this package does not include:

• No Whisper code/weights, no llama.cpp binaries, no XTTS-v2 models, no GGUF models.

Credits & Tools

• Author/Maintainer/Creative Director: Tsoxer — [e-mail]
• Code scaffolding & helper scripts: co-authored with ChatGPT
• Image assets: generated with ChatGPT (OpenAI)

Third-Party Notices (not bundled)

• OpenAI Whisper — MIT License. Source: openai/whisper.
• llama.cpp — MIT-licensed C/C++ inference project.
• XTTS-v2 (Coqui) — licensed under the Coqui Public Model License (non-commercial). You must review and comply with their terms.
• OpenHermes-2.5-Mistral-7B-GGUF Q8_0 (TheBloke) — GGUF conversions hosted on Hugging Face; follow the original/model repo licenses. Note: Q8_0 is a very large, high-quality variant and requires significant VRAM.

Third-Party Notices (bundled)

• salutations.wav by shadoWisp — used in demos; licensed CC BY 3.0. If you use it, give attribution and link the source: https://freesound.org/s/260931/

• intro.wav — generated with AudioLDM 2. Used under CC BY-NC-SA 4.0 for non-commercial, research/educational purposes only.

• Other AI-generated audio (button presses, UI/game SFX, music, etc.) — generated with TangoFlux. These clips were created using a model whose checkpoints are licensed for non-commercial research use only, subject to the Stability AI Community License (Stable Audio Open) and WavCaps’ academic-use terms. They are included here only for non-commercial, research/educational use.

This package is © 2025-2026 Tsoxer (MIT). See LICENSE.